Speech recognition method and apparatus

Abstract

 A viewpoint of a user is detected in a viewpoint detecting process, and how long the detected viewpoint has stayed in an area is determined. The obtained viewpoint and its trace is displayed on a display unit. In a recognition information controlling process, relationship between the viewpoint (in an area) and/or its movement, and recognition information (words, sentences, grammar, etc.) is obtained as weight P(). When the user pronounces a word (or sentence), the speech is inputted and A/D converted via a speech input unit. Next, in a speech recognition process, a speech recognition probability PS() is obtained. Finally, speech recognition is performed on the basis of a product of the weight P() and the speech recognition probability PS(). Accordingly, classes of the recognition information are controlled in accordance with the movement of the user's viewpoint, thereby improving the speech recognition probability and speed of recognition.

Description

[0001] The present invention relates to a speech recognition apparatus employing speech recognition method performed in accordance with the viewpoint (point where the user's eyes are focused on the display screen) of a user and an information processing apparatus using the speech recognition apparatus.

[0002] A method for improving precision of speech recognition by controlling the recognition information contained in communicative information has been proposed. The "recognition information" is managed by a recognition apparatus and includes data consisting of words, sentences, grammar, and frequency of use thereof. For example, in a calendar application, when a speech "i∂r" (phonetic symbol) is inputted if higher weight is put to a word "year" than a word "ear", the word "year" is more likely chosen. In likewise manner, the recognition information can be controlled. Further, a method for improving recognition ability of a recognition apparatus by using image information in speech recognition has been also proposed.

[0003] However, in the aforesaid speech recognition method, neither a user's visual direction nor application is taken into consideration. Thus, the recognition information is controlled in accordance with a communicative situation at best. Therefore, precision of speech recognition greatly depends on precision of control of the communicative situation. In a case where the recognition information is controlled depending upon communicative situations, it is difficult to quickly correspond to different users' manners (communicative situation) which differs from person to person, resulting in that adequate speech recognition level cannot be achieved and recognition speed cannot be improved.

[0004] The present invention has been made in consideration of the above situation, and aims in one aspect to provide a speech recognition apparatus which has improved speech recognition level and/or recognition speed by properly taking user's visual direction and application into consideration by detecting movement of a viewpoint of the user and controlling recognition information on the basis of the user's visual direction and application.

[0005] Embodiments of the present invention provide an information processing apparatus having a speech recognition apparatus which achieves aforesaid speech recognition as well as executes applications on the basis of recognition results of inputted speech.

[0006] An alternative aspect of the present invention is to improve speech recognition level by enabling the change of different classes of recognition information for speech recognition processing on the basis of a viewpoint of a user.

[0007] Yet another alternative aspect of the present invention is to improve speech recognition level by putting different weights on different classes of recognition information for speech recognition processing on the basis of a viewpoint of a user and performing the speech recognition by using the weighted recognition information.

[0008] Yet another alternative aspect of the present invention is to enable to control recognition processing on the basis of relationship between a viewpoint and positions of displayed images, such as icons of applications, on a display screen.

[0009] Yet another alternative aspect of the present invention is to enable execution of a predetermined process by selecting an application in accordance with a class of recognition information which is used to obtain recognition result in speech recognition processing performed on the basis of a viewpoint of a user, and by supplying the recognition result to the selected application.

[0010] Other features and advantages of the present invention will be apparent from the following description taken in conjunction with the accompanying drawings, in which like reference characters designate the same or similar parts throughout the figures thereof.

BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

Fig. 1 is a block diagram showing a brief configuration of an information processing apparatus according to a first embodiment;

Fig. 2 is a flowchart explaining an operation of speech recognition processing according to the first embodiment;

Fig. 3 is a graphical representation of an example of an operation according to the first embodiment;

Fig. 4 is a graphical representation showing contents of a speech input control dictionary;

Fig. 5 is a graphical representation showing contents of a dictionary for an application A;

Fig. 6 is a graphical representation showing contents of a dictionary for an application B;

Fig. 7 shows a picture on a display screen in an example of an operation according to the first embodiment;

Fig. 8 shows a picture on a display screen in an example of an operation according to the first embodiment;

Fig. 9 shows a picture on a display screen in an example of an operation according to the first embodiment;

Fig. 10 is a table showing relationship between viewpoints and weights to be applied;

Fig. 11 is a flowchart explaining an operational sequence according to the first embodiment;

Fig. 12 is a flowchart explaining the operational sequence according to the first embodiment;

Fig. 13 shows a picture on a display screen in an example of an operation according to a second embodiment;

Fig. 14 is a flowchart explaining an operational sequence according to the second embodiment;

Fig. 15 shows a picture on a display screen in an example of an operation according to a third embodiment;

Figs. 16A and 16B are graphs showing changes of weights according to a fourth embodiment;

Figs. 17A and 17B are graphs showing changes of weights with respect to time according to a fifth embodiment;

Fig. 18 shows a picture on a display screen in an example of an operation according to a sixth embodiment;

Fig. 19 is a graphical representation showing correspondence between viewpoint detection areas and weights according to a seventh embodiment;

Fig. 20 is a block diagram showing a configuration of an information processing apparatus according to the embodiments when a control program provided by a storage medium is loaded and executed;

Fig. 21 is a flowchart showing a sequence to execute each processing module provided by a storage medium; and

Fig. 22 is a memory map showing a storage situation of processing modules.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Preferred embodiments of the present invention will be described in detail in accordance with the accompanying drawings.

<First Embodiment>

[0013] Fig. 1 is a block diagram showing a brief configuration of an information processing apparatus according to a first embodiment. In Fig. 1, reference numeral 101 denotes a speech input unit including a microphone 101a and an A/D converter 101b; 102, a display unit for displaying various kinds of icons and a viewpoint (or area); 103, a viewpoint detector having a detector 103b which detects a viewpoint on the basis of an image sensed by a camera 103a or the like; and 104, a speech recognition unit for recognizing input speech using a selected class of recognition information. In the first embodiment, classes of the recognition information are weighted dictionaries, containing words, sentences, grammar, and frequency of use thereof, corresponding to different kinds of applications.

[0014] Further, reference numeral 105 denotes a recognition information controller for controlling recognition information (i.e., selecting one of the dictionaries) on the basis of a detected result of a viewpoint by a viewpoint manager 106; 106, the viewpoint manager for managing movement of a viewpoint (how long and at which point of the display unit 102 a viewpoint is fixed, and so on) detected by the viewpoint detector 103; 107, a command processing unit for performing various kinds of processes on the basis of a command detected in speech recognition.

[0015] In the configuration shown in Fig. 1, each of the speech recognition unit 104, the recognition information controller 105, the viewpoint manager 106, and the command processing unit 107 can comprise of CPU, ROM, and RAM (not shown) so as to achieve the desired functions, or a single CPU can be used for controlling the aforesaid four units to realize the same functions. When a single CPU is used, functions of each unit are obtained by executing a control program stored in the ROM 111 by CPU 110 as shown in Fig. 20. The ROM 111 stores a control program for realizing a control sequence shown in a flowchart which will be explained later. Further, RAM 112 provides a work area to the CPU 110 when it performs various kinds of control. A control program can be loaded to the RAM 112 from a floppy disk 114 or the like and executed by the CPU 110. This process will be explained later.

[0016] Next, speech recognition processing by the information processing apparatus of the first embodiment will be explained. The speech recognition processing in the first embodiment is performed by the configuration denoted by reference numerals 101 to 106 as above. Fig. 2 is a flowchart explaining an operation of speech recognition processing according to the first embodiment.

[0017] First, when a user powers on the information processing apparatus, a viewpoint of the user is detected by using the viewpoint detector 103 in viewpoint detecting process 201. Next, at which position (or an area) on the display unit 102 or its vicinity and how long the viewpoint detected in the viewpoint detecting process 201 is fixed is obtained by the viewpoint manager 106 in viewpoint managing process 202. Then, the obtained viewpoint and its trace are displayed on the display unit 102.

[0018] Thereafter, on the basis of the relationship between the viewpoint (area), its movement, and recognition information (words, sentences, grammar, and frequency of use thereof), weight P() is determined in recognition information controlling process 203. As described above, weight P() for a class of recognition information 204 is controlled in accordance with a viewpoint and its movement.

[0019] Meanwhile, when the user pronounces a word (or sentence), the speech is inputted and A/D converted via the speech input unit 101 in speech input process 205. Then, a speech recognition probability PS() is obtained in speech recognition process 206, and a product of the weight P() and the speech recognition probability PS() is obtained as a final recognition probability PT() of the word (or sentence).

[0020] In the first embodiment, a plurality of dictionaries are used as different classes of the recognition information, and one of the plurality of dictionaries is selected by putting weight "1" on a dictionary selected in accordance with the movement of the user's viewpoint, and putting weight "0" on other dictionaries. It should be noted that, in the first embodiment, weight is either "1" or "0", thus the speech recognition processing works as if different classes of recognition information are switched.

[0021] As for the aforesaid operation, it will be explained more specifically using an example. Fig. 3 is an initial state of the information processing apparatus when a user powers on the apparatus. In Fig. 3, reference numeral 301 denotes a screen of the display unit 102. Further, reference numeral 302 denotes a speech input control panel; and 305 and 308, icons of an application A (e.g., calendar) and an application B (e.g., calculator), respectively.

[0022] Further, reference numerals 303, 306, and 309 denote viewpoint detection areas for the icons 302, 305, and 308, respectively. The viewpoint detection area 303 shows the viewpoint detection area for the speech input control panel, and when a viewpoint stays inside of the viewpoint detection area 303, it is determined that the user is interested in the speech input control panel. Likewise, the viewpoint detection areas 306 and 309 are the viewpoint detection areas for the application A and the application B, respectively. It should be noted that the viewpoint detection area demarcations are not displayed on the screen 301. Further, reference numeral 304 denotes a speech input control dictionary; 307, a dictionary for the application A (calendar); and 310, a dictionary for the application B (calculator).

[0023] Reference numeral 311 denotes a current viewpoint; and 312, a trace of a viewpoint. A user's viewpoint during a predetermined past time period (e.g., for 1 sec.) is displayed. Movement of a viewpoint is sometimes stops at positions (expressed with black points in Fig. 3) and sometimes jumps (expressed with curves in Fig. 3). These kinds of movement are repeated alternatively.

[0024] Figs. 4 to 6 are graphical representations showing contents of each dictionary. Fig. 4 is a graphical representation showing contents of the speech input control dictionary 304 whose recognition words are "start" and "quit". Fig. 5 is a graphical representation showing contents of the dictionary for the application A (calendar) 307 whose recognition words (or sentences) are, for example, "1990" (nineteen-ninety), "May, 1994" (may nineteen-ninety-four), and "quit". Fig. 6 is a graphical representation showing contents of the dictionary for the application B (calculator) 310 whose recognition words (or sentences) are "3 + 6 =" (three plus six equals to) and "quit", for example.

[0025] Next, referring to Figs. 7 to 12, examples of operations will be explained. Figs. 7 to 9 show pictures on the screen in examples of operations according to the first embodiment. Further, Fig. 10 is a table showing relationship between viewpoints and weights to be applied, and Figs. 11 and 12 are flowcharts explaining an operational sequence according to the first embodiment. Step numbers which will be used below corresponds to those in the flowcharts in Figs. 11 and 12.

[0026] After the information processing apparatus is powered on, a viewpoint is detected (at a point X1 at this time), and the detected point is displayed on the screen (Fig. 7). Under a current situation, only the speech input recognition dictionary 304 is related to the viewpoint detection area for the speech input control panel 303 with weight of P(O) = 0, and neither the dictionaries for the application A nor the application B are related to their respective viewpoint detection areas (Fig. 10(1)). A detecting process for detecting whether or not the viewpoint stays in viewpoint detection areas is performed a plurality of times per second (e.g., 10 times/sec.).

[0027] Then, the viewpoint is moved to a position X2 inside the speech input control panel 302 (Fig. 7). If the viewpoint stays in the viewpoint detection area for the speech input control panel 303 more than a predetermined number of times (e.g., three times) during a predetermined time period (e.g., 1 sec.), then the viewpoint manager 106 determines that the user selected the speech input control panel 302 (at steps S10 and S11). Accordingly, the speech input recognition control dictionary 304 is related to the viewpoint recognition area for the speech input control panel 303 with weight of P(O) = 1 (Fig. 10(2), at step S12). Note that, in the process at step S12 at this point, the dictionaries of other applications are not related to the corresponding viewpoint detection areas, thus no weight is put on these dictionaries.

[0028] Thereafter, the user pronounces a word "start" while watching the speech input control panel 302 (Fig. 10(3)), and a speech recognition probability PS() for each word in the speech input control dictionary 304 which is related to the viewpoint detection area for the speech input control panel 303 is determined as follow (it is possible to employ a known method in speech recognition).




Then, the word recognition probabilities (PT() = P() × PS()) are,

[0029] By selecting a word having the greatest PT() as a recognized word, "start" is selected, thus the speech input control is to be performed. As a result, the viewpoint detection area for the application A 306 is related to the dictionary for the application A 307 with weight of P(A) = 0, and the viewpoint detection area for the application B 309 is related to the dictionary for the application B 310 with weight of P(B) = 0 (Fig. (4), at steps S14 to S16).

[0030] Then, the viewpoint is moved to a position X5 in the icon of the calendar (Fig. 7). If the viewpoint stays in the viewpoint detection area for the application A 306 more than predetermined number of times (e.g., three times) during a predetermined time period (e.g., 1 sec.), then the viewpoint manager 106 determines that the user selected the icon of the calendar 305 (at steps S10 and S20). Accordingly, the dictionary for the application A 307 is related to the viewpoint detection area for the application A 306 with weight of P(A) = 1, and words (or sentences) in the dictionary shown in Fig. 5 can be used in the speech recognition processing. Further, weight between the dictionary for the application B 310 and the viewpoint detection area for the application B 309 becomes P(B) = 0, and the weight between the speech input control dictionary 304 and the viewpoint detection area for the speech input control panel 303 becomes P(O) = 0 (Fig. 10(5), at step S21)

[0031] Then, while watching the viewpoint detection area 306 for the application A, the user pronounces "May, 1994" (may, nineteen-ninety-four) (Fig. 10(6)). Accordingly, a speech recognition probability PS() of each word (or sentence) in the dictionary for the application A 307 which is related to the viewpoint detection area for the application A 306 is determined as follow.
   PS(January, 1990) = 0.1
      ...
   PS(May, 1994) = 0.5
      ...
   PS(quit) = 0.01
Then, word recognition probabilities PT() = P() × PS() become,
   PT(January, 1990) = P(A) × PS(January, 1990) = 0.1
      ...
   PT(May, 1994) = P(A) × PS(May, 1994) = 0.5
      ...
   PT(quit) = P(A) × PS(quit) = 0.01

[0032] By selecting a word having the greatest PT(), "May, 1994" is selected as a recognized word. Accordingly, a calendar window 501 of the May 1994 is opened (Fig. 8), and the dictionary for the application A 307 is related to a viewpoint detection area for the calendar window 501 with weight of P(A) = 1" (Fig. 10(7), at steps S22 to S24). In this example, the size of the displayed area of the calendar window 501 and the viewpoint detection area for the calendar window are identical, however, they do not always have to the same size and can have different areas as for the icon of the calendar 305 and the viewpoint detection area for the calendar 306.

[0033] Next, while watching the calendar window 501, the user pronounces "December, 2001" (december, twenty-o-one) (Fig. 10(8)). Since the calendar window is a part of the viewpoint detection area for the application A, the processes step S20 to step S24 are repeated.

[0034] At step S23, a speech recognition probability PS() of each word in the dictionary for the application A 307 can be determined as follow.
   PS(January, 1990) = 0.1
      ...
   PS(December, 2001) = 0.5
      ...
   PS(quit) = 0.1
Accordingly, word recognition probabilities, PT()= P() × PS(), are,
   PT(January, 1990) = P(A) × PS(January, 1990) = 0.1
      ...
   PT(December, 2001) = P(A) × PS(December, 2001) = 0.5
      ...
   PT(quit) = P(A) × PS(quit) = 0.1

[0035] By selecting a word having the greatest PT(), "December, 2001" is selected as a recognized word, and the application A is executed in accordance with a command, or the recognized word, at step S24. Thus, a calendar window changes from the one for the May, 1994 to the one for the December, 2001. Next, while watching the viewpoint detection area for the calendar window 501, the user pronounces "quit" (Fig. 10(9)).
Accordingly, a speech recognition probability PS() for each word in the dictionary for the application A 307 is determined as follow.
   PS(January, 1990) = 0.1
      ...
   PS(quit) = 0.6
Then, word recognition probabilities PT() = P() × PS() are,
   PT(January, 1990) = P(A) × PS(January, 1990) = 0.1
      ...
   PT(quit) = P(A) × PS(quit) = 0.6

[0036] By selecting a word having the greatest PT(), "quit" is selected as a recognized word. Accordingly, the calendar application is closed as well as the calendar window 501 is closed. If the viewpoint is at a position X10 (Fig. 9) at this time, since it does not belong to the viewpoint detection area for the speech input control panel 303 nor for the application A 306 nor for the application B 309, weight becomes P(O) = P(A) = P(B) = 0 (Fig. 10(10)).

[0037] Next, the viewpoint is moved to a position X11 in the icon of the calculator 308 (Fig. 9). If the viewpoint stays in the viewpoint detection area for the application B 309 more than predetermined number of times (e.g., three times) during a predetermined time period (e.g., 1 sec.), then the viewpoint manager 106 determines that the user selected the calculator (steps S10 and S30). Accordingly, the dictionary for the application B 310 is related to the viewpoint detection area for the application B 309 with weight of P(B) = 1, and words (or sentences or arithmetic equations) in the dictionary shown in Fig. 6 can be used in the speech recognition processing. Further, the weight between the dictionary for the application A 307 and the viewpoint detection area for the application A 306 becomes P(A) = 0, and the weight between the speech input control dictionary 304 and the viewpoint detection area for the speech input control panel 303 becomes P(O) = 0 (Fig. 10(11), at step S31). In this case, when P(B) = 1, and the information processing apparatus of the first embodiment can recognize arithmetic equations and a word "quit" in the speech recognition processing.

[0038] Next, while watching the viewpoint detection area for the application B 309, the user pronounces "2 + 3 =" (two plus three equals to) (Fig. 10 (12)). Accordingly, speech recognition probabilities PS() of the arithmetic equations and a word "quit" in the dictionary for the application B 310 which is related to the viewpoint detection area for the application B 309 are determined as follow.
   PS(1 + 1 =) = 0.1
      ...
   PS(2 + 3 =) = 0.5
      ...
   PS(quit) = 0.01
Accordingly, word (arithmetic equation) recognition probabilities, PT() = P() × PS(), become,
   PT(1 + 1 =) = P(B) × PS(1 + 1 =) = 0.1
      ...
   PT(2 + 3 =) = P(B) × PS(2 + 3 =) = 0.5
      ...
   PT(quit) = P(B) × PS(quit) = 0.01

[0039] By selecting the word (arithmetic equation) having the greatest PT(), "2 + 3 =" is selected as a recognized word and the calculator application is executed. Accordingly, a calculator window is opened (not shown), and a calculated result "5" is displayed (at steps S32 to S34). The calculator window is related to the dictionary for the application B 310, similarly to the case for the calendar window. Then, while watching the viewpoint detection area for the application B 309 or the calculator window, the user pronounces "quit" (Fig. 10(13)).

[0040] Accordingly, the speech recognition probabilities PS() of the arithmetic equations and a word "quit" in the dictionary for the application B 310 which is connected to the viewpoint detection area for the application B 309 become as follow.
   PS(1 + 1 =) = 0.1
      ...
   PS(quit) = 0.6
Then, the recognition probabilities PT() = P() × PS() become,
   PT(1 + 1 =) = P(B) × PS(1 + 1 =) = 0.1
      ...
   PT(quit) = P(B) × PS(quit) = 0.6.

[0041] By selecting a word (or arithmetic equation) having the greatest PT(), "quit" is selected as a recognized word. Accordingly, the calculator application is closed and the calculator window is also closed. If the viewpoint is at a position X14 (Fig. 9) at this time, since it does not belong to the viewpoint detection area for the speech input control panel 303 nor for the application A 306 nor for the application B 309, weight becomes P(O) = P(A) = P(B) = 0 (Fig. 10(14), at step S40).

[0042] Then, the viewpoint is moved to a position X15 in the speech input control panel 302 (Fig. 9). If the viewpoint stays in the viewpoint detection area for the speech input control panel 303 more than a predetermined number of times (e.g., three times) during a predetermined time period (e.g., 1 sec.), then the viewpoint manager 106 determines that the user selected the speech input control panel 302. Accordingly, weight for each dictionary becomes P(O) = 1, P(A) = 0, P(B) = 0 (Fig. 10(15), at step S12).

[0043] Thereafter, the user pronounces a word "quit" while watching the speech input control panel 302 (Fig. 10(16)), and a speech recognition probability PS() for each word in the speech input control dictionary 304 which is related to the viewpoint detection area for the speech input control panel 303 are determined as follow.




Then, the word recognition probabilities (PT() = P() × PS()) are,

[0044] By selecting a word having the greatest PT() as a recognized word, "quit" is selected and executed. As a result, the relationship between the viewpoint detection area for the application A 306 and the dictionary for the application A 307 as well as the relationship between the viewpoint detection area 309 and the dictionary 310 both for the application B cease (Fig. 10(17), at steps S17 and S18).

[0045] According to the first embodiment as described above, classes of the recognition information are automatically switched. Further, it is possible to control the application so that, if a user pronounces a word "quit", when the user is looking at the icon of the calendar, the calendar application is terminated, and when the user is looking at the icon of the calculator, the calculator application is terminated.

[0046] Therefore, according to the first embodiment, on the basis of a viewpoint which stayed inside a viewpoint detection area set on the display screen and its vicinity, a dictionary for speech recognition (recognition information) which is related to the viewpoint detection area is selected. As a result, dictionaries used for speech recognition are switched properly, thus improving speech recognition level and speech recognition speed. Further, it is easy to operate the desired application by using speech commands, thus improving operability.

<Second Embodiment>

[0047] Next, a second embodiment will be explained. In the first embodiment, a condition for selecting an application is that a viewpoint stays in a predetermined viewpoint detection area more than a predetermined number of times in a predetermined period of time, and weight P() for a dictionary for the selected application is set to "1", whereas weight P() for a dictionary for the non-selected application is set to "0". In the second embodiment, depending upon how long the viewpoint stays inside a predetermined area, weights P() for each class of the recognition information (dictionary) which is related to an icon or a control button on the display screen are controlled.

[0048] Fig. 13 shows a picture on a display screen in an example of an operation according to the second embodiment. More specifically, Fig. 13 shows a trace of a viewpoint after a user has watched the screen 301 for 60 sec. Here, it is assumed that the viewpoint stopped in the viewpoint detection area 303 5 times, 30 times in the viewpoint detection area 306, 15 times in the viewpoint detection area 309, and 10 times in other area. Fig. 14 is a flowchart explaining an operational sequence according to the second embodiment.

[0049] In the above example shown in Fig. 13, since the result of tracing movement of a viewpoint for one minute indicates that the total number of times the viewpoint stayed in any of predetermined viewpoint detection areas is 50 (= 5 + 30 + 15), each weight is determined as follow (at steps S50 and S51).

[0050] Under this condition, assume that a user pronounce "May, 1990" (may, nineteen-ninety), and weights P() of each word (or sentence or arithmetic equation) are






Then a recognition probability PT() = P() × PS() for each word becomes,

[0051] By selecting a word (or sentence or arithmetic equation) having the greatest PT() as a recognized word, "May, 1990" is selected (at step S52 and S53). Further, an application related to the dictionary including the recognized word is specified (at step S54), and a predetermined process (e.g., to open a calendar window of May, 1990) is performed (at step S55)

[0052] According to the second embodiment as described above, by tracing movement of a viewpoint for a predetermined time period, and weight for classes of the recognition information (dictionaries related to applications) is changed on the basis of the traced result. Therefore, in a case where the viewpoint is moving and does not stay inside a single area, it is possible to stably determine weight for each dictionary. Accordingly, the user does not have to stop his viewpoint so as to be in a single area, thus increasing operability.

<Third Embodiment>

[0053] In the third embodiment, weight P() for each class of the recognition information related to an icon or a control button is controlled depending upon distances from the position where a viewpoint is to each viewpoint detection area of an application.

[0054] Fig. 15 shows a picture on a display screen in an example of an operation according to the third embodiment, and reference numeral 311 denotes a current viewpoint of a user. Distances from the viewpoint to the viewpoint detection area for the application A 306, to the viewpoint detection area for the application B 309, and to the viewpoint detection area for the speech input control panel 303 are denoted by d(A), d(B), and d(O), respectively, assuming that d(A) = 5 cm, d(B) = 20 cm, and d(O) = 25 cm.

[0055] By determining weight for a word related to the icon A, B, and the speech input control panel as




and


then, they become,

[0056] If the user pronounce "May, 1990" (may nineteen-ninety), then speech recognition probabilities PS() for words (or sentence or arithmetic equation) are obtained as,






Thus, a word recognition probability PT() = P() × PS() for each word is,

[0057] By selecting a word (or sentence or arithmetic equation) having the greatest PT(), "May 1990" is selected as a recognized word. Further, an application related to the dictionary including the recognized word is specified, and a predetermined process (e.g., opening a calendar window of May, 1990) is performed. In this embodiment, weight P() is defined as above, however, it can be defined by any functions as far as weight decreases as the distance becomes large (e.g., P(A) = exp(-d(A))).

[0058] Note that a flowchart according to the third embodiment is similar to that in the second embodiment, and it is omitted.

[0059] For instance, a current viewpoint is detected at step S50 in Fig. 14, and distances from the detected viewpoint to each of the viewpoint detection areas are calculated. Then at step S51, weight for each dictionary is calculated on the basis of the distances from the detected viewpoint to each of the viewpoint detection areas.

[0060] According to the third embodiment as described above, by detecting distances from a viewpoint and each of viewpoint detection areas, and weight for each class of the recognition information (dictionaries related to applications) is controlled on the basis of the distances. Therefore, the viewpoint does not always have to be inside of the viewpoint detection areas. In other words, the user does not have to fix his viewpoint inside of a viewpoint detection area, thus increasing operability.

<Fourth Embodiment>

[0061] Next, a forth embodiment will be now described. In the first embodiment, depending upon whether or not the viewpoint stays in a viewpoint detection area, weight for a dictionary related to the corresponding viewpoint detection area where the detected viewpoint is in is set to "1", and weights for other dictionaries are set to "0". In the fourth embodiment, a case where, when a viewpoint previously in one of viewpoint detection areas is currently out of the area, weight for each class of recognition information is not changed immediately, and keep the same weight for a fixed time period, will be explained.

[0062] Fig. 16A is a graph showing change of weight according to a fourth embodiment.

[0063] If a time t0 is set to the last minute when a current viewpoint is inside of the viewpoint detection area for the application A 306, then weight for the dictionary for the application A 307 at t0 is P(A) = 1.

[0064] As shown in Fig. 16A, if the viewpoint is outside of the viewpoint detection area 306 at time t1, the weight is not changed as long as t1 - t0 < T (e.g., T = 1 sec.), and P(A) remains 1. On the contrary, if the viewpoint is outside of the viewpoint detection area 306 at time t2 and t2 - t0 ≧ T, then the weight P(A) is 0.

[0065] Further, when a viewpoint is moved to other position outside of viewpoint detection areas, it is possible to gradually change weight for each class of the recognition information. To do so, weight for the dictionary for application P(A) is changed by following a graph shown in Fig. 16B, for example. Fig. 16B is another graph showing change of weight according to a fourth embodiment. In Fig. 16B, predetermined values (e.g., T1 = 1 sec., T2 = 3 sec.) are set to T1 and T2 (T2 > T1).

[0066] According to the fourth embodiment as described above, time allowance is set so as to correspond to quick movement of a viewpoint. Thus, a user does not have to fix his viewpoint intentionally, thereby enabling to provide an information processing apparatus of high quality and of convenient operability.

<Fifth Embodiment>

[0067] A fifth embodiment will now be explained. In the fifth embodiment, sensitivity to decide weight for classes of recognition information controlled in accordance with a viewpoint is different for each class of recognition information. A time period since a viewpoint enters an icon of an application A until the weight becomes 1 is denoted by Tin(A), and a time period since the viewpoint exits the icon of the application A until the weight becomes 0 is denoted by Tout(A). Similarly, Tin(B) and Tout(B) are denoted. Figs. 17A and 17B are graphs showing changes of weight with respect to time, where Tin(A) = 1 sec., Tout(A) = 2 sec., Tin(B) = 1 sec., and Tout(B) = 1 sec. In this case, P(A) has higher sensitivity than P(B), thus the class of the recognition information for the application (A) is more likely selected.

[0068] According to the fifth embodiment, time allowance is set so as to correspond to quick movement of a viewpoint, as in the fourth embodiment.

<Sixth Embodiment>

[0069] A sixth embodiment will be explained below. In the sixth embodiment, an example in which weight for each class of recognition information is controlled depending upon a changed viewpoint when the viewpoint is moved during recognizing speech. Fig. 18 shows a picture on a display screen in an example of an operation according to the sixth embodiment.

[0070] Assume that a user is watching an icon B, and his viewpoint is at a position S1. Under this condition, weight for each class of recognition information is,






Now, if the user pronounces "May 1990" (may nineteen-ninety), and the viewpoint just before he finished pronouncing is at a position S2, then weight for each class of recognition information changes to,

[0071] In the sixth embodiment, after speech is inputted, a viewpoint is detected in advance to speech recognition process, and weight is determined depending upon the detected viewpoint.

[0072] The processing sequence is the same as the flowchart in Fig. 14 with steps S50 and S51 moved between step S52 and step S53. At step S50, a current viewpoint is detected and, at step S51, weight corresponding to each application can be obtained on the basis of the detected viewpoint.

[0073] According to the sixth embodiment, the user has a time allowance to move his viewpoint to a desired icon before finishing the inputting speech.

<Seventh Embodiment>

[0074] In the first and second embodiments explained with reference to Figs. 7 to 9 and Fig. 13, a viewpoint is detected in a fixed time and space of each viewpoint detection area. However, the viewpoint can be determined more flexibly, such as by probability. Fig. 19 is a graphical representation showing correspondence between viewpoint detection areas and weight according to a seventh embodiment. In Fig. 19, three ranges of the viewpoint detection areas corresponding to three different weight, P(A) = 0.9, 0.5, and 0.1, are set. More specifically, the weight is set higher as the viewpoint is closer to the icon 305. Therefore, the icon which is closest to the user's viewpoint is considered as the one desired by the user. Accordingly, the probability that the icon which is considered that the user is more interested in becomes higher. Thus, there is an area allowance in detecting a viewpoint, thereby increasing operability.

[0075] Note that, in the seventh embodiment, an average of weight corresponding to areas in which the viewpoint has stayed during a predetermined time period is used as a final weight. The obtained weight is multiplied by speech recognition probability.

[0076] Further, in the aforesaid embodiments, the viewpoint is detected in image processes using a camera, however, the present invention is not limited to this, and a detecting device, such as line of sight detection glasses can be used instead of a camera.

[0077] Furthermore, in the aforesaid embodiment, recognition processing is performed on the basis of words and their recognition probabilities for the sake of simplicity. However, the recognition processing can be performed based on sentences, grammar, and recognition probabilities thereof , for example.

[0078] An object of the present invention which can be achieved by the aforesaid apparatus or the method according to the above description can also be achieved by a system storing a program according to the invention. More specifically, by attaching a storage medium to the aforesaid apparatus, the program loaded from the storage medium can constitute the new function of the invention. An example of configuration of an apparatus with the storage medium is shown in Fig. 20. CPU 110 realizes various kinds of functions by executing a control program stored in ROM 111 or RAM 112. In this example, a floppy disk 114 is inserted into a floppy disk unit 113, and a control program capable of realizing functions (functions explained with reference to the aforesaid flowchart of the speech recognition unit 104, the recognition information controller 105, and the viewpoint manager 106 shown in Fig. 1) is loaded to the RAM 112, then the CPU 110 executes the program. Note that the control program is not only provided by the floppy disk, but also by any kinds of storage medium, such as an ID card, a hard disk, and CD-ROM, or can be loaded through a network.

[0079] A structural feature of the program of the present invention is as shown in Figs. 21 and 22.

[0080] Fig. 21 is a flowchart showing an operational sequence to execute each program module provided by a storage medium. Further, Fig. 22 is a memory map showing a storage situation of program modules.

[0081] A detecting process module 401 is for controlling the viewpoint detector 103 to detect the viewpoint of a user. A changing process module 402 is for changing kinds of recognition processes in accordance with the detected viewpoint. Here, to change kinds of recognition processes, classes of recognition information (dictionary for recognition) to be used are changed in accordance with where the viewpoint stays, for instance. Alternatively, weight to be applied to each class of recognition information is changed in accordance with the viewpoint in the recognition process. A recognizing process module 403 is for recognizing speech inputted from the speech input unit 101 by following the changed recognition processes. Further, a transferring process module 404 is for transferring the recognized result to an application corresponding to the class of recognition information used by the recognizing process module.

[0082] The processes by each of aforesaid modules corresponds to control represented by flowcharts in Figs. 11, 12, and 14.

[0083] Further, the present invention can be applied to a system constituted by a plurality of devices or to an apparatus comprising a single device.

[0084] Furthermore, the invention is applicable also to a case where the invention is embodied by supplying a program to a system or apparatus. In this case, a storage medium, storing a program according to the invention, constitutes the invention. The system or apparatus installed with the program read from the storage medium realizes the functions according to the invention.

[0085] As described above, according to the present invention, user's attention and interest can be determined properly from his viewpoint, thereby it is possible to control recognition information to be used in speech recognition in accordance with the detected viewpoint. Accordingly, capabilities of the apparatus, such as speech recognition probability and speed of recognition, are improved.

[0086] As many apparently widely different embodiments of the present invention can be made without departing from the spirit and scope thereof, it is to be understood that the invention is not limited to the specific embodiments thereof except as defined in the claims.

Claims

1. A speech recognition apparatus for recognizing inputted speech characterized by comprising:
   detecting means (103) for detecting a viewpoint of a user;
   changing means (105) for changing recognition processes for speech recognition in accordance with the viewpoint detected by said detecting means; and
   recognizing means (104) for executing the speech recognition in accordance with the recognition process changed by said changing means.

2. The speech recognition apparatus according to claim 1, wherein said changing means changes classes of recognition information to be used for the speech recognition in accordance with the viewpoint detected by said detecting means.

3. The speech recognition apparatus according to claim 2, wherein said changing means sets a plurality of areas each of which is related to a different class of the recognition information, selects one of the plurality of areas in accordance with the viewpoint detected by said detecting means, and determines a class of the recognition information, related to the selected area, to be used in the speech recognition.

4. The speech recognition apparatus according to claim 3, wherein one of the plurality of areas is set for each respective image, shown on a display screen, representing an application and has a predetermined size in which the image is included.

5. The speech recognition apparatus according to claim 3, wherein said changing means detects a position where the viewpoint stays on the basis of the viewpoint detected by said detecting means and selects one of the plurality of areas which includes said position.

6. The speech recognition apparatus according to claim 1, wherein said changing means controls weight for each class of the recognition information to be used in the speech recognition in accordance with the viewpoint detected by said detecting means.

7. The speech recognition apparatus according to claim 6, wherein said changing means sets a plurality of areas each of which is related to a different class of recognition information and controls the weight for each class of the recognition information in accordance with relationship between the viewpoint detected by said detecting means and each of the plurality of areas.

8. The speech recognition apparatus according to claim 7, wherein one of the plurality of areas is set for each respective image, shown on a display screen, representing an application and has a predetermined size in which the image is included.

9. The speech recognition apparatus according to claim 7, wherein said changing means counts a number of times a viewpoint has stayed in the plurality of areas during a predetermined time period on the basis of the viewpoint detected by said detecting means, and controls weight for each class of the recognition information to be used in the speech recognition in accordance with the counted number.

10. The speech recognition apparatus according to claim 7, wherein said changing means detects a position where the viewpoint stays on the basis of the viewpoint detected by said detecting means, and controls the weight for each class of the recognition information in accordance with relationship between the position and each of the plurality of areas.

11. The speech recognition apparatus according to claim 7, wherein said changing means assigns a maximum value to weight for a class of the recognition information related to one of the plurality of areas in which the viewpoint stays, and assigns a minimum value to weight for other classes of the recognition information.

12. The speech recognition apparatus according to claim 11, wherein, in a case where the viewpoint exits from the area where it has stayed, said changing means changes the weight for the class of the recognition information related to the area in which the viewpoint had stayed to the minimum value after a predetermined time period has elapsed.

13. The speech recognition apparatus according to claim 11, wherein, in a case where the viewpoint exits from the area where it has stayed, said changing means gradually decrease the weight for the class of the recognition information related to the area in which the viewpoint had stayed toward the minimum value as time elapses.

14. The speech recognition apparatus according to claim 13, wherein rate of weight changes differ from each other depending upon different classes of the recognition information.

15. The speech recognition apparatus according to claim 11, wherein, in a case where the viewpoint enters one of the plurality of areas, said changing means gradually increases the weight for a class of the recognition information related to the area which the viewpoint entered toward the maximum value as time elapses.

16. The speech recognition apparatus according to claim 15, wherein rate of weight changes differ from each other depending upon classes of the recognition information each of which is related to one of the plurality of areas.

17. The speech recognition apparatus according to claim 1, wherein said changing means changes recognition processes on the basis of a viewpoint detected by said detecting means just before the speech recognition is performed by said recognizing means.

18. The speech recognition apparatus according to claim 7, wherein each of the plurality of areas is divided into a plurality of fields, and said changing means controls weight for each class of the recognition information in accordance with the field in which the viewpoint detected by said detecting means is.

19. A speech recognition method for recognizing inputted speech characterized by comprising:
   a detecting step (201) of detecting a viewpoint of a user;
   a changing step (203) of changing recognition processes for speech recognition in accordance with the viewpoint detected in said detecting step; and
   a recognizing step (206) of executing the speech recognition in accordance with the recognition process changed in said changing step.

20. The speech recognition method according to claim 19, wherein, in said changing step, classes of recognition information to be used for the speech recognition are changed in accordance with the viewpoint detected in said detecting step.

21. The speech recognition method according to claim 20, wherein, in said changing step, a plurality of areas each of which is related to a different class of the recognition information are set, and one of the plurality of areas is selected in accordance with the viewpoint detected in said detecting step, and a class of the recognition information, related to the selected area, to be used in the speech recognition is determined.

22. The speech recognition method according to claim 21, wherein one of the plurality of areas is set for each respective image, shown on a display screen, representing an application and has a predetermined size in which the image is included.

23. The speech recognition method according to claim 21, wherein, in said changing step, a position where the viewpoint stays is detected on the basis of the viewpoint detected in said detecting step and one of the plurality of areas which includes said position is selected.

24. The speech recognition method according to claim 19, wherein, in said changing step, weight for each class of recognition information to be used in the recognition process is controlled in accordance with the viewpoint detected in said detecting step.

25. The speech recognition method according to claim 24, wherein, in said changing step, a plurality of areas each of which is related to a different class of recognition information are set and the weight for each class of the recognition information is controlled in accordance with relationship between the viewpoint detected in said detecting step and each of the plurality of areas.

26. The speech recognition method according to claim 25, wherein one of the plurality of areas is set for each respective image, shown on a display screen, representing an application and has a predetermined size in which the image is included.

27. The speech recognition method according to claim 25, wherein, in said changing step, a number of times a viewpoint has stayed in the plurality of areas is counted during a predetermined time period on the basis of the viewpoint detected in said detecting step, and weight for each class of the recognition information to be used in the speech recognition is controlled in accordance with the counted number.

28. The speech recognition method according to claim 25, wherein, in said changing step, a position where the viewpoint stays is detected on the basis of the viewpoint detected in said detecting step, and the weight for each class of the recognition information is controlled in accordance with relationship between the position and each of the plurality of areas.

29. The speech recognition method according to claim 25, wherein, in said changing step, a maximum value is assigned to weight for a class of the recognition information related to one of the plurality of areas in which the viewpoint stays, and a minimum value is assigned to weight for other classes of the recognition information.

30. The speech recognition method according to claim 29, wherein, in a case where the viewpoint exits from the area where it has stayed, the weight for the class of the recognition information related to the area in which the viewpoint had stayed is changed to the minimum value, after a predetermined time period has elapsed, in said changing step.

31. The speech recognition method according to claim 29, wherein, in a case where the viewpoint exits from the area where it has stayed, the weight for the class of the recognition information related to the area in which the viewpoint had stayed is gradually decreased, toward the minimum value as time elapses, in said changing step.

32. The speech recognition method according to claim 31, wherein rate of weight changes differ from each other depending upon different classes of the recognition information.

33. The speech recognition method according to claim 29, wherein, in a case where the viewpoint enters one of the plurality of areas, the weight for a class of the recognition information related to the area which the viewpoint entered is gradually increased, toward the maximum value as time elapses, in said changing step.

34. The speech recognition method according to claim 33, wherein rate of weight changes differ from each other depending upon classes of the recognition information each of which is related to one of the plurality of areas.

35. The speech recognition method according to claim 19, wherein, in said changing step, recognition processes on the basis of a viewpoint detected in said detecting step are changed just before the speech recognition is performed in said recognizing step.

36. The speech recognition method according to claim 25, wherein each of the plurality of areas is divided into a plurality of fields, and, in said changing step, weight for each class of the recognition information is controlled in accordance with the field in which the viewpoint detected in said detecting step is.

37. An information processing apparatus for executing an application characterized by comprising:
   detecting means (103) for detecting a viewpoint of a user;
   changing means (105) for changing recognition processes for speech recognition in accordance with the viewpoint detected by said detecting means;
   recognizing means (104) for executing the speech recognition in accordance with the recognition process changed by said changing means;
   selecting means for selecting an application to be executed; and
   executing means for executing the application on the basis of a speech recognition result by said recognizing means.

38. The speech recognition apparatus according to claim 37, wherein said changing means selects a class of recognition information to be used for the speech recognition in accordance with the viewpoint detected by said detecting means, and said selecting means selects an application in accordance with the class of the recognition information selected by said changing means.

39. The speech recognition apparatus according to claim 37, wherein said changing means controls weight for each class of the recognition information to be used in the speech recognition in accordance with the viewpoint detected by said detecting means, and said selecting means selects an application in accordance with the class of the recognition information to be used for obtaining the recognition result by said recognizing means.

40. The information processing apparatus according to claim 37, further comprising display means for displaying images representing each of a plurality of applications,
   wherein said changing means sets a plurality of areas each of which includes each of the plurality of images and changes recognition processes in accordance with relationship between the viewpoint detected by said detecting means and each of the plurality of areas.

41. A computer program product comprising a computer usable medium having computer readable program code means embodied in said medium for speech recognition processing, said product including:
   first computer readable program code means (401) for a detecting step of detecting a viewpoint of a user;
   second computer readable program code means (402) for a changing step of changing recognition processes for speech recognition in accordance with the viewpoint detected in said detecting step; and
   third computer readable program code means (403) for a recognizing step of executing the speech recognition in accordance with the recognition process changed in said changing step.

42. A computer program product according to claim 41, wherein said second computer readable program code means changes classes of recognition information to be used for the speech recognition in accordance with the viewpoint detected in the detecting step.

43. A computer program product according to claim 41, wherein said second computer readable program code means controls weight of each class of the recognition information to be used in the speech recognition in accordance with the viewpoint detected in the detecting step.

44. A computer program product according to claim 41 further comprising fourth computer readable program code means for selecting one of the plurality of applications in accordance with a class of the recognition information to be used in the recognising step and transmitting a recognised result in said recognition process to the selected application.

45. An apparatus according to any of claims 1 to 18, wherein said detecting means (103), said changing means (105) and said recognising means (104) are implemented by program control of a computer apparatus.

46. A recording medium having the program for control according to claim 45 recorded thereon.

47. Use of view direction measurements to determine contact in speech recognition.

48. A method of speech recognition using a system having a display, characterised by changing the recognition process in dependence upon where in the display the user is looking.

Drawing